Parameter Estimation in a Land-Surface Model Using Atmospheric Data Assimilation: Finding Distributions for Use in an Ensemble Prediction System

To account for uncertainty in parameters of a land-surface model (such as surface albedo, surface emissivity, soil volumetric heat capacity) this study constructs and interprets distributions of these parameters in land-surface parameterization schemes in a column model that is assimilating near-surface atmospheric observations. The goals are to determine which parameters are most sensitive to observations and whether they demonstrate distributions that could be exploited in an ensemble forecasting system. TKE-based and bulk boundary layer schemes, and a simple land surface model, are implemented in a column model to simulate the behavior of a mesoscale NWP model. Observations from the Atmospheric Radiation Measurement (ARM) program Southern Great Plains Central Facility near Lamont, OK, are assimilated with an ensemble filter. Parameters in the scheme are treated as stochastic variables, and their distributions are estimated by including them in the state vector during assimilation. The combined assimilation/column system has different sensitivities than a free-running column system. Some of those distributions can be easily characterized and sampled to form a stochastic-dynamic ensemble forecasting system, while others are not easily established and may prove more effective when fixed. Additional comments will be made regarding the suitability of ensemble data assimilation for identifying systematic model error in PBL and land-surface models.